Skip to main content

Revision-Aware Implant Placement: A Computational Planning Framework for Quantifying the Reconstructive Options That Remain After Implant Loss

Paper type
Computational modeling methodology and retrospective validation protocol
Status
Original Work · Working Paper
Contents
  1. 1Introduction
  2. 2Background and Current Scientific Evidence
  3. 3Research Gap
  4. 4Research Objective and Central Hypothesis
  5. 5Scientific and Theoretical Framework
  6. 6Methodology
  7. 7Analytical Development
  8. 8Discussion
  9. 9Clinical Significance
  10. 10Limitations
  11. 11Future Validation
  12. 12Conclusion

Abstract

Background. Virtual implant planning optimizes the position of a fixture for the restoration it will carry and for the anatomy it must avoid. It does not ask what the site will permit if the implant is later lost. Systematic reviews report lower survival for implants placed at previously failed sites than for first placements, and the dimensional changes that follow implant removal are poorly documented.

Objective. To define a quantitative measure of the reconstructive options that would remain after loss of a planned implant, to embed that measure in multi-objective planning, and to specify the studies needed to test whether it has any practical value.

Framework. A planned implant pose is evaluated under a set of failure scenarios. Each scenario removes bone according to a failure-mode template and an explantation template and then applies a parameterized remodeling envelope. The residual jaw is searched for feasible replacement trajectories under the same safety and prosthetic constraints used for the original plan. The reconstructive flexibility index (RFI) is the scenario-weighted fraction of the original feasible pose space that survives. Plans are compared on a Pareto front of prosthetic fit, bone support, and RFI, with the first two given lexicographic priority.

Proposed validation. Four stages are described: algorithm verification on synthetic phantoms; a retrospective paired comparison of clinician plans and revision-aware plans on archived cone-beam computed tomography; calibration of the defect model against archived pre-placement and post-removal scans; and a blinded expert-panel assessment.

Central hypothesis. Among plans that are equivalent on prosthetic and safety criteria, RFI varies by a clinically meaningful amount, so that flexibility can often be preserved at no cost to the first attempt.

Conclusions. The framework is a proposal. An analytic reduction shows that its leverage depends on one empirical assumption, that bone loss after implant removal depends nonlinearly on residual wall thickness, and that assumption has not been tested. No results are reported.

Keywords: dental implants; computer-assisted treatment planning; implant failure; reimplantation; cone-beam computed tomography; multi-objective optimization; alveolar bone loss

1. Introduction

Implant planning has been organized around the restoration for three decades. The fixture is positioned where the planned crown or framework needs it, the ridge is augmented when it does not offer bone in that position, and the plan is transferred to the mouth with a guide or a navigation system [1–3]. Cone-beam computed tomography (CBCT) and planning software made this approach routine, and recent work on automated planning has kept the same objective function while delegating the search to an algorithm [4–7].

The objective function has a horizon of one implant lifetime. It is evaluated at the moment of placement and, implicitly, for as long as the implant survives. Long-term survival is high: a meta-analysis of prospective studies estimated 10-year implant-level survival at 96.4% (95% CI 95.2%–97.5%), with a prediction interval of 91.5%–99.4% [8], and a review of studies with at least 10 years of follow-up reported a cumulative mean of 94.6% [9]. Those figures also mean that several implants in every hundred are lost, that the proportion is higher in identifiable subgroups, and that the denominator grows every year as the implanted population ages [10–12].

When an implant is lost, the clinician returns to a site that the first implant has altered. Reimplantation is feasible but less successful than first placement. One systematic review found 88.84% survival for second attempts (390 of 439 implants; range across studies 71%–94.6%) and 74.19% for third attempts (23 of 31) [13]. Another reported an overall survival of 86.3% after retreatment and concluded that the rate is lower than that generally reported for initial placement [14]. Individual cohorts have reported 83.5%, 73%, and 71% [15–17]. A further meta-analysis addressed the same question [18]. None of these studies relates the outcome of the second attempt to how the first implant was positioned.

That missing link is the subject of this manuscript. Two plans can be indistinguishable on every criterion a planning system reports (angulation relative to the prosthetic axis, distance to the inferior alveolar canal, buccal wall thickness above a threshold) and still differ in what they would leave behind. One may consume the only corridor of bone in which a 4-mm implant can be placed; the other may leave a second corridor intact. One may use the full available height; the other may leave apical reserve for a longer replacement. Planning software does not display this difference because nothing in its objective function measures it.

A measure of it is proposed here. The reconstructive flexibility index (RFI) expresses how much of a site’s feasible implant-pose space would survive the loss and removal of a planned implant. The index is computed by simulating failure scenarios on the planning CBCT and re-running the same feasibility search that produced the original plan. The remainder of this manuscript defines the index, identifies the assumptions on which it rests, and describes a staged program to test whether it changes plans in ways clinicians would accept. No claim is made that it improves outcomes. The framework has not been implemented, and whether flexibility that is preserved on a scan translates into better retreatment is an empirical question that only prospective data can answer.

2. Background and Current Scientific Evidence

What prosthetically driven planning optimizes

The restoration-driven concept placed the implant according to the planned restoration and treated the site as something to be developed to receive it [1]. For the anterior maxilla, this was refined into mesiodistal, orofacial, and apicocoronal comfort and danger zones [2]. Static guided surgery transfers such plans with a mean error of 1.2 mm at the entry point, 1.4 mm at the apex, and 3.5° in angulation in a meta-analysis of clinical studies [19]; sources of error and their mitigation have been reviewed [20]. CBCT use in planning is governed by guidelines that weigh diagnostic yield against radiation dose [4, 21].

Automation has followed the same logic. A scoping review of artificial intelligence in presurgical planning found that most systems address segmentation of anatomical structures and that automated positioning remains at an early stage [5]; deep learning approaches to planning on CBCT have been described [6], and a recent systematic review surveys the field [7]. The patent literature shows the same objective. One granted United States patent claims a method that prepares information for automated or semi-automated planning from jaw image data by generating tooth-setup and jaw models, detecting restorative and anatomical elements, selecting implant-suitable zones, and generating, comparing, and refining plans [22]. Other filings describe positioning a dental component relative to boundary surfaces of the restoration and positioning implants with regard to bone density and nerve paths [23, 24]. In the records retrieved, the quantity being optimized is the quality of the current placement. No claim or publication was found that scores a plan by the options it leaves for a subsequent one, but the patent review was limited to index records and, for one patent, the first claim.

Outcomes of reimplantation

The evidence on reimplantation is retrospective and small. The systematic review by Zhou and colleagues included five retrospective cohorts and two case series covering 396 patients [13]. Oh and colleagues noted that the data were insufficient to evaluate the effect of patient or treatment characteristics on retreatment survival, and observed higher survival with rough-surfaced than with smooth-surfaced replacement implants [14]. In a matched comparison, first-time implants survived at 94% and implants replacing failed ones at 73% [16]. Earlier cohorts reported 83.5% and 71% [15, 17].

Three interpretations of the lower survival are compatible with these data. The first is patient-level: the same host factors that led to the first failure persist. The second is site-level and biological: the healed failure site differs in vascularity or cellularity. The third is site-level and geometric: the replacement implant is placed in a compromised volume, with less native bone contact, thinner walls, or a forced change in position. Published studies cannot separate them, because none recorded the geometry of the residual site in a standardized way. Zhou and colleagues list bone quality, density, and vertical and horizontal dimensions among site-related risk indicators without quantifying them [13]. The geometric interpretation is the only one a planning algorithm can act on, and it is the one this framework addresses.

Explantation and what it removes

A systematic review identified five explantation techniques: reverse torque, trephines, burs, piezosurgery, and laser-assisted removal [25]. The techniques differ in how much bone they sacrifice. Reverse-torque removal breaks the bone–implant interface and withdraws the fixture along its own axis; trephine removal cuts an annulus of bone around it. Reviews and technique reports concur that the conservative approach should be attempted first and that trephines and burs carry risks to adjacent structures, including sinus perforation in at least one report [25–29]. For planning purposes, the relevant point is that the defect produced by explantation is at least the envelope of the implant and may be considerably larger, and that which technique will be possible depends partly on features fixed at placement: implant diameter relative to ridge width, proximity to adjacent roots, and connection geometry.

Dimensional change after implant removal

The remodeling that follows tooth extraction is well characterized. In human re-entry studies, horizontal bone loss of 29%–63% and vertical loss of 11%–22% occurred within six months [30]; major changes take place during the first year [31]; and the buccal wall is disproportionately affected in the canine model [32]. No comparable body of data exists for the site of a removed implant. A recent review of the management and sequelae of implant removal states that knowledge of the biological events and dimensional changes after removal is lacking and that recommendations rest largely on empirical speculation [33]. An extraction socket is lined by bundle bone that depended on a periodontal ligament; a failed implant site is lined either by native bone that has lost its interface or by the walls of a peri-implantitis defect. Extrapolating socket data to implant sites is therefore an assumption, and this framework treats it as one.

The pattern of bone loss before removal

Late failures seldom leave the bone as the implant found it. In a population-based analysis, peri-implantitis-associated bone loss followed a non-linear, accelerating pattern, with onset early after loading in most affected implants [12]. The 2017 World Workshop characterized peri-implantitis lesions and their progression [34, 35]. The defect at the time of removal is thus a function of failure mode and of time to diagnosis, neither of which is known at planning. A revision-aware plan cannot predict the defect. It can only ask how sensitive the remaining options are to a plausible range of defects, which is a question about robustness.

Summary of the evidence base

Prosthetically driven planning is established and commercially mature. Reimplantation survival is lower than first-placement survival in every review identified, but the reasons are not resolved. Explantation techniques are described, and their relative invasiveness is understood qualitatively. Dimensional change after implant removal is essentially unmeasured. No published planning method was found that evaluates a plan by its consequences for a later one.

3. Research Gap

The gap has three parts. First, there is no defined quantity that expresses the reconstructive options remaining at a site after loss of an implant in a given position; clinicians reason about this informally, when they do so at all. Second, there is no method for computing such a quantity from the planning data set. Third, the empirical inputs that such a method needs (defect geometry by failure mode and explantation technique, and remodeling after removal) are missing from the literature, so any method must be built to expose its dependence on them.

A fourth issue is conceptual and must be faced before the others. Because first-placement survival is high, the expected benefit of preserving options is small for the average patient. A revision-aware criterion is defensible only if it is either costless to the first attempt or confined to patients in whom the probability of failure is materially elevated. Whether costless choices exist in ordinary sites is unknown and is the first thing a study should establish.

4. Research Objective and Central Hypothesis

The objective is to specify and prepare for testing a planning criterion that quantifies future reconstructive flexibility.

The central hypothesis is that, across archived CBCT data sets of sites planned for single implants, the set of plans equivalent to the clinician’s plan on prespecified prosthetic and safety metrics will, in a majority of sites, contain plans whose RFI differs from that of the clinician’s plan by at least a prespecified minimum important difference. A corollary is that revision-aware plans selected from that equivalent set will preserve more feasible reimplantation trajectories than the clinician plans did.

Two secondary hypotheses concern the model itself. The residual feasible space predicted by the defect model will agree with the residual feasible space measured on post-removal scans of the same sites to within prespecified limits of agreement. And experienced clinicians, blinded to the origin of each plan, will rate revision-aware plans as acceptable for the first attempt no less often than conventional plans.

The hypothesis would be refuted if RFI rarely varies among equivalent plans, if it varies only when the remodeling parameters are set to values later shown to be unrealistic, or if clinicians judge the revision-aware plans inferior for the primary restoration.

5. Scientific and Theoretical Framework

Pose space and feasibility

Let Ω denote the bone volume of the site segmented from CBCT, with a density surrogate ρ(x) and a set of protected structures A (inferior alveolar canal, mental foramen, sinus and nasal floors, adjacent roots, and the outer cortical surfaces). An implant pose is p=(e,a,L,D): an entry point e on the crest, a unit axis a, a length L, and a diameter D drawn from a finite catalog. The pose occupies a volume V(p).

A pose is feasible when it satisfies every constraint in a set that mirrors current planning rules: a minimum distance m_j to each protected structure j∈A; a minimum circumferential bone thickness t_(min) over a specified coronal length; an angular deviation from the prosthetic axis no greater than θ_(max); and emergence within the restorative envelope E. The feasible set is

F(Ω)={ p:g_k(p;Ω,A,E)≤0 for all k }  (1)

and its size is measured by μ(F), the number of feasible poses on a fixed sampling lattice of entry points, axis directions, and catalog sizes. The lattice makes μ a count of distinguishable trajectories. Its resolution is a numerical parameter whose influence must be shown to vanish on refinement.

Failure scenarios and the residual site

A failure scenario s combines a failure mode with an explantation technique. A minimal scenario set is proposed (Table 1). Each scenario defines a defect operator Δ_s that enlarges the implant volume and a remodeling operator R_ϕ that removes further bone as a function of local wall thickness, with parameters ϕ. The residual site after loss of an implant at pose p under scenario s is

Ω′_s(p)=R_ϕ​( Ω\Δ_s​(V(p)))  (2)

The remodeling operator is where the framework is weakest. In the absence of data from implant removal sites, it is parameterized here as a thickness-dependent loss: a wall whose thickness after defect creation is below a critical value t_c is lost over a height h_c, and thicker walls lose a fraction λ of their thickness. The values of t_c, h_c, and λ are not known for failed implant sites. They are to be varied over ranges bounded provisionally by the post-extraction literature [30–32] and then calibrated in Stage 3 of the validation program.

Table 1. Proposed failure scenarios. Defect dimensions are parameters to be calibrated; no values are asserted.
ScenarioFailure modeBone loss before removalExplantation templateRationale
S1Early failure, no integrationNone beyond the osteotomyWithdrawal along axisLower bound on damage
S2Late biological failure, contained defectCircumferential crater of depth d and width wReverse torqueTypical peri-implantitis morphology, conservative removal [25, 34]
S3Late biological failure, wall lossCrater with loss of one cortical wallReverse torqueDefect open to buccal or lingual
S4Mechanical failure of an integrated implantNoneTrephine annulus of thickness wt to depth LtFractured fixture not retrievable by torque [26]
S5Integrated implant removed for malpositionNoneBur or piezoelectric window on one aspectRemoval with intact interface [33]

The reconstructive flexibility index

For a plan p and scenario weights w_s summing to one,

RFI(p)=∑_s^​ w_s (μ​(F(Ω′_s(p))))/(μ​(F(Ω)))  (3)

RFI lies between zero (no feasible replacement trajectory in any scenario) and a value near one (the site retains almost all of its original options). Because conditional probabilities of each failure mode are not available by site, equal weights are the default and the per-scenario terms are always reported alongside the aggregate.

Two refinements make the index clinically interpretable. A tiered version computes the feasible set at three levels of intervention: reimplantation into native bone without grafting (tier 0), reimplantation with simultaneous contained-defect grafting, in which the thickness constraint is relaxed where the defect has intact walls (tier 1), and staged reconstruction (tier 2, feasible by definition). The quantity of interest is the lowest tier at which any trajectory is feasible, together with the count at that tier. A catalog-restricted version asks whether the replacement must be wider or longer than the original, since “rescue” by a larger implant is the common clinical recourse and is the first option to disappear when the original implant already used the largest size that fit.

Planning as constrained multi-objective selection

Let P(p) measure prosthetic fit (for example, the negative of a weighted sum of angular and positional deviation from the prosthetic axis and emergence point) and B(p) measure bone support (for example, density-weighted contact area). Revision-aware planning is the problem

maxp∈F(Ω) (P(p), B(p), RFI(p))  (4)

solved as a Pareto problem; an evolutionary algorithm such as NSGA-II is one suitable solver [36], although exhaustive enumeration on the lattice is feasible for single sites. Scalarizing the three objectives is not proposed. The clinical rule is lexicographic: RFI is used to choose among plans whose P and B fall within equivalence margins ε_P and ε_B of the best attainable values. Setting those margins is a clinical judgment that the framework makes explicit and that the expert-panel stage is designed to inform.

Execution error

A plan is not executed exactly. With guided-surgery errors on the order of a millimeter at the entry point and a few degrees in angulation [19], the pose achieved is a random perturbation of the pose planned. The index should therefore be evaluated as an expectation over an execution-error distribution π(p′∣p):

RFI(p)=∫RFI(p′) π(p′∣p) dp′  (5)

approximated by Monte Carlo sampling. A plan whose flexibility collapses under a one-millimeter perturbation is not robustly revision-aware, whatever its nominal score.

6. Methodology

Stage 1: implementation and verification

Segmentation. Bone, teeth, and the inferior alveolar canal are segmented from CBCT with an open platform [37] or a validated automated tool, and reviewed by a clinician. CBCT gray values are not calibrated Hounsfield units; the density surrogate is used only for relative comparison within a scan, and this limitation is carried through to the definition of B(p).

Feasibility search. Signed Euclidean distance transforms to each protected structure and to the outer bone surface are precomputed, so that each constraint in equation (1) reduces to sampling distance fields along the implant axis and surface. The default lattice uses entry points at 0.5-mm spacing within the restorative envelope, axis directions at 2° increments within a cone of half-angle θ_(max), and a catalog of diameters and lengths representative of a commercial system. Each of these is a parameter of the method.

Verification. Before any patient data are used, the code is verified on synthetic phantoms (rectangular and trapezoidal ridges with a cylindrical canal) for which μ(F) can be computed analytically. Lattice-convergence studies report the change in RFI on successive refinements; the working resolution is the coarsest at which RFI changes by less than a prespecified tolerance. Verification, validation, and uncertainty quantification follow the structure of ASME V&V 40 and the corresponding FDA guidance on computational model credibility [38, 39].

Stage 2: retrospective paired comparison

Design. A retrospective, cross-sectional, paired in silico study on archived planning data, reported according to STROBE [40].

Data. Archived CBCT volumes and the corresponding clinician-generated plans for single-implant sites, sampled in three strata: posterior mandible, posterior maxilla, and anterior maxilla. Inclusion requires a scan of diagnostic quality with a field of view covering the site and adjacent teeth, an archived plan, and a digital wax-up or equivalent definition of the restoration. Sites planned with simultaneous sinus elevation or major augmentation are excluded from the primary analysis, because the feasible set then depends on the graft. No scan is acquired for research [21].

Comparators. The clinician plan p_c is the reference. The revision-aware plan p_r is the member of the equivalence set {p:P(p)≥P^*−ε_P, B(p)≥B^*−ε_B} with the highest expected RFI. A third plan, p_o, the optimizer’s choice with the RFI term switched off, separates the effect of revision awareness from the effect of algorithmic optimization as such.

Outcomes. The primary outcome is the paired difference in expected RFI between p_r and p_c. Secondary outcomes are the proportion of sites in which the equivalence set contains any plan exceeding p_c by the minimum important difference; the lowest feasible tier under each scenario; prosthetic deviation and minimum safety margins for each plan; and the proportion of sites in which the clinician plan already lies on the Pareto front.

Sample size. No estimate of the between-site standard deviation of the paired difference exists. A pilot of 30 sites (10 per stratum) will estimate σ_d. The main-study sample size per stratum then follows from the standard paired formula n=((z_(1−α/2)+z_(1−β)) σ_d/δ)^2, with δ the minimum important difference fixed before the pilot is unblinded. No value for σ_d is assumed here.

Analysis. Mixed-effects models with a random intercept for patient account for multiple sites per patient. Stratum is a fixed effect. Sensitivity analyses vary the remodeling parameters ϕ over their full provisional ranges, the scenario weights from equal to single-scenario, the equivalence margins, and the execution-error distribution. The headline result is the proportion of the parameter space in which the primary conclusion holds, and a tornado analysis identifies the parameters that drive it.

Stage 3: calibration of the defect model

Design. A retrospective agreement study. Cases are sites with a CBCT before implant placement and a second CBCT acquired for clinical reasons after implant removal and healing. Such cases are uncommon, and multicenter pooling will be necessary.

Procedure. Scans are rigidly registered on stable anatomy. The achieved implant pose is recovered from any intermediate imaging or from the operative record. The observed residual site is segmented from the second scan. The model is run forward from the first scan with the documented failure mode and explantation technique, and the remodeling parameters are fitted by minimizing the discrepancy between predicted and observed residual bone surfaces on a training subset.

Outcomes. Agreement between predicted and observed residual bone volume and between predicted and observed μ(F(Ω′)) on a held-out subset, by Bland–Altman limits of agreement and intraclass correlation [41, 42]. This stage also yields the first systematic description of dimensional change after implant removal by failure mode, which is of independent value given the absence of such data [33].

Stage 4: blinded expert assessment

Board-certified surgeons and prosthodontists review pairs of plans (p_c and p_r) for a subset of sites in random order, blinded to origin. They rate each plan’s acceptability for the primary restoration on an ordinal scale and indicate a preference. After unblinding the RFI information, they indicate whether it would change their choice. The outcome is non-inferiority of acceptability for p_r with a prespecified margin, and the proportion of decisions changed. Inter-rater reliability is reported.

Bias, confounding, and reproducibility

Selection bias in Stage 2 is limited by consecutive sampling from the archive within strata. In Stage 3, cases with post-removal CBCT are a selected group, probably enriched for difficult sites; the calibrated parameters must be reported with that caveat. Segmentation is performed blind to plan origin. All code, the sampling lattice, the implant catalog, the constraint set, and the parameter ranges are to be released with the manuscript so that the index can be recomputed by others; without that, RFI is not a defined quantity.

Ethical considerations

All stages use archived images and require institutional review board approval or a determination of exemption, with de-identification of DICOM data. No patient is exposed to additional radiation. Stage 4 involves clinicians as participants and requires their consent.

7. Analytical Development

A two-dimensional reduction

A reduced model shows what the index is sensitive to. Consider a bucco-lingual cross-section of a ridge of uniform width W and an implant of diameter D whose center lies at distance x from the buccal surface. The buccal and lingual walls have thickness

t_B=x−D/2,  t_L=W−x−D/2  (6)

Suppose failure produces a circumferential defect of radial extent δ, and that remodeling removes r(t) from a wall of post-defect thickness t. The residual walls are t_B′=t_B−δ−r(t_B−δ) and likewise for t_L′. A replacement implant of diameter D′≥D+2δ, which is what engaging native bone without grafting requires, can be centered anywhere along an interval of length

ℓ=(t_B′+t_L′+D+2δ)−D′−2 t_(min)  (7)

provided each residual wall alone is at least t_(min) when the replacement is concentric. If r is a constant or a fixed fraction, t_B′+t_L′ depends on x only through the constraint that neither wall is negative, and ℓ is essentially independent of where the first implant was placed. Position then has no leverage on flexibility in this cross-section.

Leverage appears when r is nonlinear in thickness. Let walls thinner than t_c be lost entirely and thicker walls lose a fraction λ:

r(t)=(t,t<t_cλt,t≥t_c)  (8)

For a ridge in which a centered implant leaves both walls just above t_(min) but below t_c+δ, a late failure removes both walls and tier 0 reimplantation is impossible. Shifting the implant lingually by a fraction of a millimeter, within the prosthetic equivalence margin, can keep the lingual wall above the critical thickness and leave a one-wall defect that is graftable at tier 1. The total bone is the same. What changes is whether one wall survives.

This reduction makes a testable statement. The sensitivity of RFI to bucco-lingual position is governed by the nonlinearity of r(t). If Stage 3 shows that wall loss after implant removal is proportional to thickness, with no threshold, the bucco-lingual component of the framework loses most of its rationale. The threshold behavior has experimental support for extraction sockets, where the thin buccal wall is preferentially lost [32], but that is not evidence about implant removal sites.

Apical reserve

In the vertical dimension, the argument does not depend on remodeling. If the available height above a protected structure is H and the safety margin is m, a first implant of length L leaves an apical reserve

ρ_a=H−m−L  (9)

A replacement implant that gains stability from previously unengaged apical bone requires ρ_a>0. A plan that uses the longest implant that fits sets ρ_a=0 and removes that recourse. Whether the stability gained by the last two millimeters of length in the first implant is worth the loss of apical reserve is a trade-off between B(p) and RFI that the Pareto front displays and that current planning does not make visible.

Mesiodistal corridors and multi-implant plans

For multiple implants, the index generalizes by evaluating loss of each implant in turn and, for full-arch cases, by counting the fallback sites that remain usable for a replacement fixture given the positions of the survivors and the geometry of the existing framework. Two plans with the same anteroposterior spread may differ in whether any posterior bone remains unoccupied on each side. This extension connects to the structural question of how much mechanical reserve a full-arch prosthesis retains after loss of a support, which is a separate problem requiring separate methods.

When the index should carry weight

Let π_f be the probability that the implant is lost within the planning horizon. The expected contribution of a flexibility difference ΔRFI to any utility is proportional to π_f⋅ΔRFI. With π_f of a few percent over 10 years in general populations [8], the product is small, and no sacrifice of first-attempt quality can be justified by it. This is the formal reason for the lexicographic rule. The weight rises with π_f: in sites of previous failure, where the next attempt has an observed failure probability on the order of 11%–29% across reviews and cohorts [13, 16, 17], and in patients with recognized risk indicators [11, 34]. A reasonable implementation would display RFI in all cases and allow it to influence selection beyond the equivalence set only when the clinician has explicitly judged the risk to be elevated.

8. Discussion

The proposal changes the question a plan answers. Prosthetically driven planning asks where the implant should be for this restoration. Revision-aware planning adds a second question, asked only among acceptable answers to the first: which of these positions leaves the patient best placed if the implant is lost. It is a tie-breaking rule with a defined metric.

Several objections deserve a direct response. The first is that failure is too rare to plan for. The lexicographic structure concedes this for the average case; the framework matters only if equivalence sets are large enough that tie-breaking is frequent, and Stage 2 is designed to find out. If clinician plans turn out to be near-unique optima in most sites, the framework has little to offer and the study will show that.

The second objection is that the defect cannot be predicted. That is correct, and the framework does not attempt prediction for an individual. It evaluates a plan against a fixed panel of scenarios, in the way a structure is checked against standard load cases without knowing which will occur. What matters is whether the ranking of plans is stable across the scenario panel and the parameter ranges. If rankings invert when remodeling parameters move within plausible bounds, the index is not usable until Stage 3 narrows those bounds.

The third is that attention to flexibility could encourage undersized or malpositioned first implants, harming the outcome that matters most. The equivalence margins are the safeguard, and the expert-panel stage tests it. A related concern is automation bias: a displayed score invites optimization of the score. The index should be presented with its per-scenario components and its sensitivity range, never as a single number without context.

The fourth is that lower retreatment survival may be mostly a patient-level phenomenon, in which case geometry is beside the point. The reimplantation literature cannot distinguish the explanations [13, 14]. If Stage 3 data accumulate, the association between residual-site geometry and second-attempt outcome can be examined directly, with adjustment for patient factors. A null association would undercut the clinical motivation even if the index proved computable and stable.

The framework has a side benefit that does not depend on its primary hypothesis. Implementing it requires a standardized description of defects after implant removal, and collecting that description addresses a documented gap [33]. It also requires explicit equivalence margins for prosthetic and safety criteria, which are implicit in practice and seldom stated.

There is a competing way to reach the same goal: a learned model trained on outcomes of reimplantation, predicting retreatment success from the first plan. The data to train such a model do not exist and, given the event rate, will not exist soon. A mechanistic, geometry-based index can be defined and tested now, and can later serve as a feature in a learned model if the data mature [5, 43].

9. Clinical Significance

If the hypothesis is supported, the immediate clinical product is a display in planning software: for the current plan, the number of feasible replacement trajectories that would remain under each scenario, and whether an alternative within tolerance preserves more. This would inform choices that are currently made without that information: implant length when height is limited, bucco-lingual position in narrow ridges, the decision to use the widest implant a ridge accepts, and site selection in multi-implant plans.

The framework is most relevant to patients in whom failure is more likely and retreatment more consequential: sites of previous failure, patients with a history of periodontitis or other recognized risk indicators [34], and younger patients, for whom the planning horizon exceeds the follow-up of any survival study. It also bears on consent. A discussion of what would be done if an implant fails is part of informed consent, and it is more concrete when the plan has been evaluated for that contingency.

None of this is a claim about patient outcomes. The chain from a higher index to a better retreatment result has several untested links.

10. Limitations

The remodeling operator is not supported by data from implant removal sites; its form and parameter ranges are borrowed from extraction-socket studies, which may not transfer. The scenario panel is a simplification of a continuum of failure presentations, and equal weighting is arbitrary. The feasibility constraints encode one set of planning rules; other rules give other feasible sets, and the index is meaningful only relative to a stated rule set. CBCT limits the method: gray values are not quantitative density, thin cortical plates are subject to partial-volume error at typical voxel sizes, and metal artifacts degrade post-removal scans less than post-placement scans but still affect registration. Soft tissue, which often determines whether a reconstruction is practical, is not represented. Systemic and behavioral determinants of retreatment success are outside the model. The retrospective designs cannot show clinical benefit. Stage 3 depends on a scarce type of case and may be underpowered for calibration by failure mode. Finally, the literature review was conducted from abstracts and index records, and the patent review did not include a professional search or claim-by-claim reading.

11. Future Validation

Beyond the four stages described, three further steps are needed. A prospective registry should record, for every implant removal at participating centers, the failure mode, explantation technique, and standardized imaging where clinically indicated, so that the remodeling parameters can be estimated with known precision. A prospective cohort of reimplantations should test whether residual-site geometry, summarized by tier and trajectory count, predicts second-attempt survival independently of patient factors. Only if both succeed would a randomized comparison of conventional and revision-aware planning be justified, and it would need to be confined to elevated-risk sites to accrue events. Independent replication of the Stage 2 analysis by a group using different software and a different rule set is necessary to show that the index is not an artifact of one implementation.

12. Conclusion

A plan for a dental implant can be scored for what it would leave behind. The reconstructive flexibility index defined here does this by simulating loss of the planned implant under a panel of failure scenarios and counting the feasible replacement trajectories that remain. The index is computable from data already acquired for planning, and it can be used as a tie-breaker without compromising the first attempt. Its usefulness depends on two things that are not yet known: whether plans that are equivalent for the first restoration differ meaningfully in flexibility, and whether bone loss after implant removal has the thickness-dependent character that gives implant position its leverage. The validation program is designed to answer both. Until it does, the framework is a hypothesis about planning and should be read as one.

Cite this paper

Jain S, Gupta A. Revision-Aware Implant Placement: A Computational Planning Framework for Quantifying the Reconstructive Options That Remain After Implant Loss. Working paper. Center for Implant Dentistry; 2026. Available from: https://bayareaimplantdentistry.com/clinician-research-and-publications/revision-aware-implant-placement

References

  1. Garber DA, Belser UC. Restoration-driven implant placement with restoration-generated site development. Compend Contin Educ Dent. 1995;16(8):796, 798-802, 804. PMID: 8620398 (opens in a new tab).
  2. Buser D, Martin W, Belser UC. Optimizing esthetics for implant restorations in the anterior maxilla: anatomic and surgical considerations. Int J Oral Maxillofac Implants. 2004;19 Suppl:43-61. PMID: 15635945 (opens in a new tab).
  3. D’haese J, Ackhurst J, Wismeijer D, De Bruyn H, Tahmaseb A. Current state of the art of computer-guided implant surgery. Periodontol 2000. 2017;73(1):121-133. doi:10.1111/prd.12175 (opens in a new tab). PMID: 28000275 (opens in a new tab).
  4. Jacobs R, Salmon B, Codari M, Hassan B, Bornstein MM. Cone beam computed tomography in implant dentistry: recommendations for clinical use. BMC Oral Health. 2018;18(1):88. doi:10.1186/s12903-018-0523-5 (opens in a new tab). PMID: 29764458 (opens in a new tab).
  5. Elgarba BM, Fontenele RC, Tarce M, Jacobs R. Artificial intelligence serving pre-surgical digital implant planning: A scoping review. J Dent. 2024;143:104862. doi:10.1016/j.jdent.2024.104862 (opens in a new tab). PMID: 38336018 (opens in a new tab).
  6. Kurt Bayrakdar S, Orhan K, Bayrakdar IS, Bilgir E, Ezhov M, Gusarev M, et al. A deep learning approach for dental implant planning in cone-beam computed tomography images. BMC Med Imaging. 2021;21(1):86. doi:10.1186/s12880-021-00618-z (opens in a new tab). PMID: 34011314 (opens in a new tab).
  7. Zaww K, Abbas H, Vanegas Sáenz JR, Hong G. AI-driven innovations for dental implant treatment planning: A systematic review. J Prosthodont Res. 2026;70(2):192-205. doi:10.2186/jpr.JPR_D_24_00338 (opens in a new tab). PMID: 40967790 (opens in a new tab).
  8. Howe MS, Keys W, Richards D. Long-term (10-year) dental implant survival: A systematic review and sensitivity meta-analysis. J Dent. 2019;84:9-21. doi:10.1016/j.jdent.2019.03.008 (opens in a new tab). PMID: 30904559 (opens in a new tab).
  9. Moraschini V, Poubel LA, Ferreira VF, Barboza Edos S. Evaluation of survival and success rates of dental implants reported in longitudinal studies with a follow-up period of at least 10 years: a systematic review. Int J Oral Maxillofac Surg. 2015;44(3):377-88. doi:10.1016/j.ijom.2014.10.023 (opens in a new tab). PMID: 25467739 (opens in a new tab).
  10. Esposito M, Hirsch JM, Lekholm U, Thomsen P. Biological factors contributing to failures of osseointegrated oral implants. (I). Success criteria and epidemiology. Eur J Oral Sci. 1998;106(1):527-51. doi:10.1046/j.0909-8836..t01-2-.x (opens in a new tab). PMID: 9527353 (opens in a new tab).
  11. Chrcanovic BR, Kisch J, Albrektsson T, Wennerberg A. Factors Influencing Early Dental Implant Failures. J Dent Res. 2016;95(9):995-1002. doi:10.1177/0022034516646098 (opens in a new tab). PMID: 27146701 (opens in a new tab).
  12. Derks J, Schaller D, Håkansson J, Wennström JL, Tomasi C, Berglundh T. Peri-implantitis - onset and pattern of progression. J Clin Periodontol. 2016;43(4):383-8. doi:10.1111/jcpe.12535 (opens in a new tab). PMID: 26900869 (opens in a new tab).
  13. Zhou W, Wang F, Monje A, Elnayef B, Huang W, Wu Y. Feasibility of Dental Implant Replacement in Failed Sites: A Systematic Review. Int J Oral Maxillofac Implants. 2016;31(3):535-45. doi:10.11607/jomi.4312 (opens in a new tab). PMID: 27183062 (opens in a new tab).
  14. Oh SL, Shiau HJ, Reynolds MA. Survival of dental implants at sites after implant failure: A systematic review. J Prosthet Dent. 2020;123(1):54-60. doi:10.1016/j.prosdent.2018.11.007 (opens in a new tab). PMID: 31027959 (opens in a new tab).
  15. Machtei EE, Mahler D, Oettinger-Barak O, Zuabi O, Horwitz J. Dental implants placed in previously failed sites: survival rate and factors affecting the outcome. Clin Oral Implants Res. 2008;19(3):259-64. doi:10.1111/j.1600-0501.2007.01466.x (opens in a new tab). PMID: 18177430 (opens in a new tab).
  16. Chrcanovic BR, Kisch J, Albrektsson T, Wennerberg A. Survival of dental implants placed in sites of previously failed implants. Clin Oral Implants Res. 2017;28(11):1348-1353. doi:10.1111/clr.12992 (opens in a new tab). PMID: 27743398 (opens in a new tab).
  17. Grossmann Y, Levin L. Success and survival of single dental implants placed in sites of previously failed implants. J Periodontol. 2007;78(9):1670-4. doi:10.1902/jop.2007.060516 (opens in a new tab). PMID: 17760534 (opens in a new tab).
  18. Gomes GH, Misawa MYO, Fernandes C, Pannuti CM, Saraiva L, Huynh-Ba G, et al. A systematic review and meta-analysis of the survival rate of implants placed in previously failed sites. Braz Oral Res. 2018;32:e27. doi:10.1590/1807-3107bor-2018.vol32.0027 (opens in a new tab). PMID: 29723338 (opens in a new tab).
  19. Tahmaseb A, Wu V, Wismeijer D, Coucke W, Evans C. The accuracy of static computer-aided implant surgery: A systematic review and meta-analysis. Clin Oral Implants Res. 2018;29 Suppl 16:416-435. doi:10.1111/clr.13346 (opens in a new tab). PMID: 30328191 (opens in a new tab).
  20. Chackartchi T, Romanos GE, Parkanyi L, Schwarz F, Sculean A. Reducing errors in guided implant surgery to optimize treatment outcomes. Periodontol 2000. 2022;88(1):64-72. doi:10.1111/prd.12411 (opens in a new tab). PMID: 35103317 (opens in a new tab).
  21. Bornstein MM, Scarfe WC, Vaughn VM, Jacobs R. Cone beam computed tomography in implant dentistry: a systematic review focusing on guidelines, indications, and radiation dose risks. Int J Oral Maxillofac Implants. 2014;29 Suppl:55-77. doi:10.11607/jomi.2014suppl.g1.4 (opens in a new tab). PMID: 24660190 (opens in a new tab).
  22. Materialise Dental NV, assignee. Method for (semi-) automatic dental implant planning. United States patent US 8,582,870 B2 (divisional; parent PCT/EP2006/007008 filed 2006 Jul 17).
  23. Nobel Biocare Services AG, applicant. Method and system for dental planning. European patent application EP 2 156 350 A1. Priority 2007 May 25; published 2010 Feb 24 (related WO 2008/145293 A2).
  24. Nobel Biocare Services AG, assignee. Method and system for dental planning and production. United States patent US 10,206,757 B2 (related publications US 2011/0008751 A1; WO 2008/083857 A1).
  25. Roy M, Loutan L, Garavaglia G, Hashim D. Removal of osseointegrated dental implants: a systematic review of explantation techniques. Clin Oral Investig. 2020;24(1):47-60. doi:10.1007/s00784-019-03127-0 (opens in a new tab). PMID: 31729576 (opens in a new tab).
  26. Stajčić Z, Stojčev Stajčić LJ, Kalanović M, Đinić A, Divekar N, Rodić M. Removal of dental implants: review of five different techniques. Int J Oral Maxillofac Surg. 2016;45(5):641-8. doi:10.1016/j.ijom.2015.11.003 (opens in a new tab). PMID: 26688293 (opens in a new tab).
  27. Anitua E, Orive G. A new approach for atraumatic implant explantation and immediate implant installation. Oral Surg Oral Med Oral Pathol Oral Radiol. 2012;113(3):e19-25. doi:10.1016/j.tripleo.2011.06.035 (opens in a new tab). PMID: 22669153 (opens in a new tab).
  28. Solderer A, Al-Jazrawi A, Sahrmann P, Jung R, Attin T, Schmidlin PR. Removal of failed dental implants revisited: Questions and answers. Clin Exp Dent Res. 2019;5(6):712-724. doi:10.1002/cre2.234 (opens in a new tab). PMID: 31890309 (opens in a new tab).
  29. Tafuri G, Santilli M, Manciocchi E, Rexhepi I, D’Addazio G, Caputi S, et al. A systematic review on removal of osseointegrated implants: un update. BMC Oral Health. 2023;23(1):756. doi:10.1186/s12903-023-03438-5 (opens in a new tab). PMID: 37833674 (opens in a new tab).
  30. Tan WL, Wong TL, Wong MC, Lang NP. A systematic review of post-extractional alveolar hard and soft tissue dimensional changes in humans. Clin Oral Implants Res. 2012;23 Suppl 5:1-21. doi:10.1111/j.1600-0501.2011.02375.x (opens in a new tab). PMID: 22211303 (opens in a new tab).
  31. Schropp L, Wenzel A, Kostopoulos L, Karring T. Bone healing and soft tissue contour changes following single-tooth extraction: a clinical and radiographic 12-month prospective study. Int J Periodontics Restorative Dent. 2003;23(4):313-23. PMID: 12956475 (opens in a new tab).
  32. Araújo MG, Lindhe J. Dimensional ridge alterations following tooth extraction. An experimental study in the dog. J Clin Periodontol. 2005;32(2):212-8. doi:10.1111/j.1600-051X.2005.00642.x (opens in a new tab). PMID: 15691354 (opens in a new tab).
  33. Monje A, Nart J. Management and sequelae of dental implant removal. Periodontol 2000. 2022;88(1):182-200. doi:10.1111/prd.12418 (opens in a new tab). PMID: 35103326 (opens in a new tab).
  34. Schwarz F, Derks J, Monje A, Wang HL. Peri-implantitis. J Periodontol. 2018;89 Suppl 1:S267-S290. doi:10.1002/JPER.16-0350 (opens in a new tab). PMID: 29926957 (opens in a new tab).
  35. Berglundh T, Armitage G, Araujo MG, Avila-Ortiz G, Blanco J, Camargo PM, et al. Peri-implant diseases and conditions: Consensus report of workgroup 4 of the 2017 World Workshop on the Classification of Periodontal and Peri-Implant Diseases and Conditions. J Clin Periodontol. 2018;45 Suppl 20:S286-S291. doi:10.1111/jcpe.12957 (opens in a new tab). PMID: 29926491 (opens in a new tab).
  36. Deb K, Pratap A, Agarwal S, Meyarivan T. A fast and elitist multiobjective genetic algorithm: NSGA-II. IEEE Trans. Evol. Computat.. 2002;6(2):182-197. doi:10.1109/4235.996017 (opens in a new tab).
  37. Fedorov A, Beichel R, Kalpathy-Cramer J, Finet J, Fillion-Robin JC, Pujol S, et al. 3D Slicer as an image computing platform for the Quantitative Imaging Network. Magn Reson Imaging. 2012;30(9):1323-41. doi:10.1016/j.mri.2012.05.001 (opens in a new tab). PMID: 22770690 (opens in a new tab).
  38. American Society of Mechanical Engineers. Assessing credibility of computational modeling through verification and validation: application to medical devices. ASME V&V 40-2018. New York: ASME; 2018.
  39. U.S. Food and Drug Administration. Assessing the credibility of computational modeling and simulation in medical device submissions: guidance for industry and Food and Drug Administration staff. Silver Spring (MD): FDA; 2023.
  40. von Elm E, Altman DG, Egger M, Pocock SJ, Gøtzsche PC, Vandenbroucke JP. The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement: guidelines for reporting observational studies. Lancet. 2007;370(9596):1453-7. doi:10.1016/S0140-6736(07)61602-X (opens in a new tab). PMID: 18064739 (opens in a new tab).
  41. Bland JM, Altman DG. Statistical methods for assessing agreement between two methods of clinical measurement. Lancet. 1986;1(8476):307-10. PMID: 2868172 (opens in a new tab).
  42. Koo TK, Li MY. A Guideline of Selecting and Reporting Intraclass Correlation Coefficients for Reliability Research. J Chiropr Med. 2016;15(2):155-63. doi:10.1016/j.jcm.2016.02.012 (opens in a new tab). PMID: 27330520 (opens in a new tab).
  43. Collins GS, Moons KGM, Dhiman P, Riley RD, Beam AL, Van Calster B, et al. TRIPOD+AI statement: updated guidance for reporting clinical prediction models that use regression or machine learning methods. BMJ. 2024;385:e078378. doi:10.1136/bmj-2023-078378 (opens in a new tab). PMID: 38626948 (opens in a new tab).
  44. U.S. Food and Drug Administration. Product classification: product code LLZ (medical image management and processing system), 21 CFR 892.2050; 510(k) summaries K152078 (3Shape Implant Studio) and K103723 (SICAT Implant). Silver Spring (MD): FDA.